nemi cost analysis: optical versus copper...
TRANSCRIPT
0
NEMI Cost Analysis: Optical Versus Copper
Backplanes
Presented on behalf of project members by Adam Singer
ECWC / APEXFebruary 22, 2005
1
OE TIG ProjectsOE TIG Projects
Optoelectronics Substrates Project
Leader: Jack Fisher (consultant)[email protected]
Sub-groups
PCB cost model: Adam [email protected]
Optical system: Peter [email protected]
Optical materials: Bruce [email protected]
2
AgendaAgenda
• Background / reason for iNEMI project• Updated goal• Status report
– Cost modeling– System descriptions– Optical connector challenges– Optical waveguide technologies
• Conclusions
3
OE Trends & DriversOE Trends & Drivers
• Starting to see OE equipment revenue growth, e.g. access & FTTP,PONs
• Traffic continues to grow 50-100% pa, triple play (data, VoIP, video), retail, file swapping...
• Distributed systems (1000s servers) require high speed interconnects• Copper can support up to 10G over 1m, 40G in some cases, at
increased cost: mtrls, line spacing, signal conditioning...• Emerging OE technologies: reconfigurable/active systems, polymer
waveguides & substrates, Si optical emission & modulation (Intel), optical band gaps...
• Although fiber interconnects will be continue to be used, optical functions, initially transmission, will co-exist with electrical in the printed wiring board
• COST reduction of pkg, assembly & test, remains significant driver
4
Market StatusMarket Status
Courtesy of Prismark, September 2002
5
Current Interconnect BottleneckCurrent Interconnect Bottleneck
• Increasing chip-performance leads to problems with conventional electrical intrasystem interconnection technologies (Siemens SBS C-LAB, 2002):
– Increasing electromagnetic radiation– Limited bandwidth due to high frequency skin effect– High pin-count through limited bandwidth– Increasing signal integrity and timing problems– Increased data rate leads to disproportional increased costs per
bit• Transmission speeds reach their physical limits when power
consumption is increased and there is no obvious increase in thepropagation speed
• It is the ability of optical interconnections to operate at high data rates (>2.5 GHz) with superior signal integrity that makes them an exploitable alternative
6
JIEP Optical Packaging RoadmapJIEP Optical Packaging Roadmap
Source: Japan Institute of Electronics Packaging & SEMI
7
Copper RoadmapCopper Roadmap
Teradyne connectors HDM VHDM & VHDM L-series VHDM-HSD Gbk GBx
ERNI connectors ERmet Ermet Ermet ZD ERmetZD Ermet Zero XT
Tyco connectors HM-ZD Z-Pack HM-ZD Z-Pack HM-ZD
MultiGig RT-1 MultiGig RT-2 MultiGig RT-3
FCI Metral 2000 AirMax VS AirMax VS AirMax VS AirMax
Metral 4000
Fujitsu FCN-261Z00x FCN260D
Winchester Xcell SIP-1000 I platform
Molex Molex is a Teradyne licensee
Taps required
Velio GigaCore2
Shared Bus ?Point to Point ?Multi point ? Sub type?
Generic Copper Backplane Bandwidth Technology Roadmap
PTFE / Ceramic, Df = 0.002 - 0.0009FR-4, Df = 0.020 PPO / CE, Df = 0.015-0.008
BT / APPE, Df = 0.010
PTFE, Df = 0.009 - 0.003
1 Gbps ─ 2.5 GBPS ─ 3.125 Gbps ─ 4.0 Gbps ─ 6.125 Gbps ─ 10 Gbps - 40Gbps
Materials
PCB Technology
Connector Launch Design
Transmission Line Design
Differential pair, length, type (Surface microstrip, embedded microstrip, stripline, edge coupled, broadside coupled), location in stack Single transmission line, Length management
Processes Standard processes Backdrilling ─ Dual density drilling?
Bus Architecture
Receiver / Transmitter Signal Conditioning Chip
Set
Pad in via or PTH, micro vias, buried vias
Decrease PTH diameter, Remove non-functional pads, Increase anti-pad diameter (Clearance ring)
Connector Technology
Via Design
GigaCore
None Required One Two Multiple
8
Optoelectronics ConceptsOptoelectronics Concepts
Photons• Zero Rest mass• FxD > 1014 Hz x 100km• Boson => Mult. Signals• OE Conversion needed
Best for long distance,high speed
Electrons• 9.11 x 10-31kg• FxD > 1010 Hz x 0.001km• Fermion: One Signal• No OE Conversion
Best for very short distance(<5 meters) at moderate tohigh speed
9
Optoelectronics Substrates ProjectOptoelectronics Substrates Project
Original Purpose Statements:• This purpose of this group is to addresses the implementation of
optical and optoelectronic technologies in printed wiring boards(PWB’s) used for packaging, or for other applications.
The areas to be addressed include:• An understanding of the drivers and the constraints of producing
optoelectronic PWB’s including cost analysis and tradeoffs.• Design considerations for materials used for PWB fabrication and
assembly, including material properties• Manufacturability of waveguides and the integration of waveguides in
PWB’s• Component mounting and interconnecting structures• Performance and testing of waveguides and connector attachments
10
Optical Optical BackplaneBackplane CostCost
Cost-performance is the key driver; we need an industry metricto compare optical vs. Cu-based, e.g. $/(Gb/s/channel/m)
Crossover zone:changeover will not be
immediate, but will rangedepending on issues includingcost sensitivity, reliability, and
design limitations
Bandwidth x Distance
Rel
ativ
e C
ost Copper
PCB
Optical PCB
20042000
11
Optoelectronics Substrates ProjectOptoelectronics Substrates Project
Current Statement:• Initial investigation (2002) indicated that the OEMs were not planning
to use optoelectronics in their next generation machines• The OEMs were all working on internal analysis of optoelectronic
solutions and were all interested in participating in a NEMI technology analysis activity
• There are numerous estimates of how far copper can be pushed to increase data rates. The estimates range from 2.5 Gb/s to 40 Gb/s.
• It was decided to do a business analysis of copper vs. optoelectronics
• The product to be analyzed will be a communications industry backplane, e.g. Terabit router
12
Optoelectronics Substrates ProjectOptoelectronics Substrates Project
• Project Objectives:– Develop cost models, including assembly and test, of a high
performance “product” (black box emulator)– Copper and optical backplane versions– Run cost models for various speeds and system configurations,
using appropriate sets of components and technologies
• Project Benefits:– Information exchange between suppliers (component, technology..),
assemblers and system OEMs– Define industry relevant cost-performance metric(s)– Identify technology gaps and barriers to implementation – Roadmaps and cost models will be available for industry use
• Meeting Objectives:– Report work-in-progress– Promote awareness, attract new participants
13
Optoelectronics Substrates ProjectOptoelectronics Substrates Project
• Project started in June 2003, in open/forming stage• Bi-weekly telecons• Very good participation
– Several OEMs, fabricators, CMs, material suppliers• First pass output of copper backplane model run and validated by
3 fabricators• Draft electrical technology road map,1-10(40) Gbps, includes
connectors, signal conditioning, chip-sets and high performance substrates
• Starting to develop connector input to the model• Discussing the optoelectronic technology alternatives
– Fiber, Waveguide, Polymer• Developing list of optical attributes and performance issues• Draft optical roadmap by June
14
Cost Modeling MethodCost Modeling Method
• Based on Technical Cost Modeling, as developed by IBIS Associates / MIT
• Activities Based Costing, plus engineering relationships– Assembly cycle time = f(design, mtrls, equipment)
Design
Materials Processing
Design
Materials
Machine Limits
Cycle Time Calculation
Equipment & Related
Inputs Cost output
Labor
Building
Utilities
15
Electrical Backplane Cost ModelElectrical Backplane Cost Model
Context• North American facility• Equipment assumes max 24 inch width conveyors• Yields reflective of 8 mil line / space• Drilling reflects 20 mil minimum diameter• Medium throughput (350Ksqft/yr top surface)
Product• 32 metal layers, no buried vias• FR-4, 24x20 inch panel, 20x18 inch finished board• 5,000 drilled holes per board• ~1-3 Gbps performance
16
Back Plane Cost ModelBack Plane Cost Model
17
Strategy for System Cost ModelsStrategy for System Cost Models
• Seek input from relevant experts:– Technology and hardware selection, identify gaps (white paper)– Issues & concerns, e.g. manufacturing yield, reliability– Requirements for performance, application, future system
(IP, confidentiality issues)• Define detailed sets of attributes, electrical & optical• Draft roadmaps: understand which and how the attributes change
with performance• Define 1st generation black box system (top down approach)• Choose logical set(s) of the key attributes (80/20)• Obtain cost data, build initial cost model• Verify and improve model with actual costs, expert input• Run what if sensitivities: speed, no. lines, hardware variation,
yields, …• Repeat for 2nd generation system
18
Electrical Backplane Attributes 1Electrical Backplane Attributes 1--40G 40G (Partial)(Partial)
• Receiver/Transmitter signal conditioning chip set:– Bus type: multipoint, point-to-point– Signaling: single-ended, differential, CML NRZ, 4PAM– Standard: DDR2/3/4, LVDS, XAUI, CEI-6G-SR/LR, CEI-11G-SR– Equalisation: 400 mV emphasis, DFE 5/8 tap, duobinary
• PCB Technology:– Materials: FR4, PPO/CE, FR4013, BT/APPE, PTFE– Loss: Df = 0.02 - 0.002– Processes: standard, backdrilling, dual density drilling– Transmission line: single-ended, loosely coupled, broadside– Distance: up to 1m– Via/connector launch: PTH, pad in via, micro, buried, differential anti-
pad
• Connector Technology:– Several vendor types
19
Optical Backplane Attributes 1Optical Backplane Attributes 1--40G 40G (Partial)(Partial)
• Receiver/Transmitter:– Link loss budget: 0.5 dB per channel– Source/modulation: < 5G VCSEL, <10G F-P(int), >10G DFB (ext)– Channel density: transceivers, VCSEL / PD arrays– Multiplex: serial (TDM), parallel (12-72 channels), CWDM
• Light coupling:– Type: butt coupled, index matched, lens, mirror/grating, free-space
• Optical carrier technology: – Fiber: single, ribbon, flex/laminated (shuffle), MM or SM– Waveguide: surface, embedded, organic, silica, PLC– Impact on PCB design rules, electrical layer reduction
• Connector:– Coupling losses: tolerance, cleanliness (contamination has critical impact
on yield & costs, but hard to quantify)
• Assembly & Test
20
ETHERNET (Layer2/3 Switch)ETHERNET (Layer2/3 Switch)WB #1 Electrical, XAUI over the backplaneWB #1 Electrical, XAUI over the backplane
Electrical Line Card
BP Interface connector C
XFP 10 Gb/s
Ethernet MACNP/TM PHY
10 Gb/s
XFI
3.125 Gb/s
XAUI
622 Mb/s
SPI4.2
3.125 Gb/s
XAUI
Switch Card
BP Interface connector C
Layer 2/3 Switch Fabric
3.125 Gb/s
XAUI
FIC: Fabric Interface ChipNP/TM: Network Processor/Traffic ManagerMAC: Medium Access Controller .PHY: Physical Interface
21
ETHERNET (Layer2/3 Switch)ETHERNET (Layer2/3 Switch)WB #1 Electrical, 10G Native over the backplaneWB #1 Electrical, 10G Native over the backplane
Electrical Line Card
BP Interface connector B
XFP 10 Gb/s
Ethernet MACNP/TM PHY
10 Gb/s
XFI
Nom 10 Gb/s
10GbE
622 Mb/s
SPI4.2
Nom 10 Gb/s
10GbE
Switch Card
BP Interface connector B
Layer 2/3 Switch Fabric
Nom 10 Gb/s
10GbE
FIC: Fabric Interface ChipNP/TM: Network Processor/Traffic ManagerMAC: Medium Access ControllerPHY: Physical Interface
22
ETHERNET (Layer2/3 Switch)ETHERNET (Layer2/3 Switch)BB #1 Optical, XAUI rate over the backplaneBB #1 Optical, XAUI rate over the backplane
Electrical Line Card
SNAP-12 or POP-4
XFP 10 Gb/s
Ethernet MACNP/TM PHY
10 Gb/s
XFI
3.125 Gb/s
XAUI
622 Mb/s
SPI4.2
3.125 Gb/s
XAUI
Switch Card
SNAP-12 or POP-4
Layer 2/3 Switch Fabric
3.125 Gb/s
XAUI
FIC: Fabric Interface ChipNP/TM: Network Processor/Traffic ManagerMAC: Medium Access Controller .PHY: Physical Interface
23
ETHERNET (Layer2/3 Switch)ETHERNET (Layer2/3 Switch)BB #1 Optical, 10G Native over the backplaneBB #1 Optical, 10G Native over the backplane
Electrical Line CardBP Interface SNAP-12 or POP-4 (10G)
XFP 10 Gb/s
Ethernet MACNP/TM PHY
10 Gb/s
XFI
Nom 10 Gb/s
10GbE
622 Mb/s
SPI4.2
Nom 10 Gb/s
10GbE?
Switch Card
Layer 2/3 Switch Fabric
Nom 10 Gb/s
10GbE?BP Interface SNAP-12 or POP-4 (10G)
FIC: Fabric Interface ChipNP/TM: Network Processor/Traffic ManagerMAC: Medium Access ControllerPHY: Physical Interface
24
Future Optical Black BoxesFuture Optical Black Boxes
• Optical Black Box #2:– 40G implementation of BB#1, and/or fully populated
frame (3 ATCA shelves, 16 cards per shelf), and/or full mesh system (high channel count) TBD
– Fiber based– Possibly use CWDM optical multiplexed signals within
the fabric to overcome space restrictions• Optical Black Box #3:
– Fiber Flexplane version of BB#2 (TBD)• Optical Black Box #4:
– Organic embedded waveguide in backplane, version of BB#2
27
OECB Technology ChallengesOECB Technology Challenges
n x 4 flipchip VCSEL array Optical layer, with
embedded MM waveguides
Optical
underfill
Optical via Multi-layer FR4 board Optical edge
connectorChoice of Glass or Polymer optical layers
Choice of embedded or overlaid waveguides
45o mirror
Choice of assembled or integrated mirrors
28
Coupling ChallengeCoupling Challenge
Ideal process route is to combine mirror into
board during PCB fabrication, for direct active
component attach.
Limited alignment tolerance stack-up.
Low cost assembly achievable.
Source: Siemens SBS C-Labs
VCSEL
Mirror
Integrated Mirror ?
Several research avenues involving the insertion
of mirrors or carriers.
Large alignment tolerance stack-up.
High cost issues due to small parts, and tight
alignment tolerances.
Assembled Mirror ?
VCSELMirror
Carrier
29
Optical Material CharacteristicsOptical Material CharacteristicsCoupling Applications Producer Status
Methods (3)Optical WaveguidesPolymer
Deposited Fiber + Free Space Backplanes, General interconnect R&D
Photoimaged
Direct fiber/component, I/O mirror arrays, Array connectors,
single/stacked layers
Stand alone flex or board/substrate interconnects-chip, component, functions, links, fully connectorized,
stacked or single layer. Optical CrossLinks
Custom Prototying & Production
Embossed Press Fiber + Surface (4) WDM, Splitters, Couplers, ADM OptoFoil - Fraunhofer Institute Production Roll Fiber + Surface (4) Ribbon Cables, Backplanes 3M, Promerus Production
Trench & Fill Fiber + Surface DWDM, VOA, ADM, Splitters, Couplers Shipley, DuPont Photonics, NTT, Neophotonics R&D
Micromolded Fiber/F.S./Surface (4) Light Pipes, Backplanes, Passive Interconnect Promerus Production
Molded Fiber + Free Space Light Pipes, Backplanes ?
Inorganic
Silica on Silicon Fiber end DWDM, ADM, AWG Neophotonics, Symmorphics Production
Polysilicon Fiber end Chip Switches, Modulators Intel Research
Hybrids (6) Fiber end VOA, ADM Lightwave Microsystems, Neophotonics Production
Optical FiberEmbedded Fiber Glass Fiber + Surface Backplanes, Lightpipes Hitachi Chemical (Wire Wrap) R&D Polymer Fiber + Surface Backplanes, Lightpipes Northrup-Grunman R&D
30
Optical Material Characteristics (2)Optical Material Characteristics (2)Field Size WG Mode Waveguide NRE Cost Cost
in. 830 nm 1550 nm Type Structure Pitch (2) $/Layer (5) $/ft2/Layer (5)Optical WaveguidesPolymer
Deposited 12 X 18 0.1 0.5 Rib Multimode Coarse 1K 5 - 10
Photoimaged
12 X 18, 13" reel to
reel 0.08 0.7 Rib & EmbeddedSingle Mode &
Multimode Hyperfine 100 to~2500 30-50
Embossed Press Wafer (1) 0.1 0.1 Embedded Single/Multi Fine 10K 1,500 Roll 36 X 36 0.1 0.3 Rib & Embedded Single/Multi Fine 20K 5 - 10
Trench & Fill Wafer 0.1 0.1 Embedded Single/Multi Fine 10K 1,500
Micromolded 36 X 36 0.1 0.3 Rib & Embedded Single/Multi Fine 12K 10
Molded 24 X 24 0.1 0.5 Rib Multimode Coarse >50K 5
Inorganic
Silica on Silicon Wafer < 0.1 < 0.1 Rib & Embedded Single/Multi Fine 10K 1,500
Polysilicon Wafer ? ? Embedded Single Mode Fine 15K 3000
Hybrids (6) Wafer < 0.1 < 0.1 Rib & Embedded Single Fine 15K 1,700
Optical FiberEmbedded Fiber Glass 18 X 24 < 0.1 < 0.1 Embedded Single/Multi Coarse 5K 50 Polymer 18 X 24 < 0.1 < 0.1 Embedded Multimode Coarse 5K 30
Attenuation, dB/cm
31
Next StepsNext Steps
• Run multiple copper sensitivities– Materials, via qty. and types,– Layers counts, etc.
• Develop connector data base– Cost, type, application
• Add signal conditioning, bus chip-sets, ASICs– May need resource assistance
• Complete optical attributes and optical technology roadmap• Determine the copper / opto architectures to be modeled for high
performance– Difficult because it is often proprietary information
32
ConclusionsConclusions
• Performance of copper backplanes will continue to improve, at increased cost to achieve signal integrity
• Benefits of optical outweigh cost for critical applications, e,g, avionics, clock distribution in high perf systems
• Future generation, high bandwidth telecom/office systems likely to have optical distribution
• The problems of assembling and handling fiber/connectors are drivers for integrating optical waveguides in the PWB
• Improved coupling and light-turning technologies are required• The initial choices of optical technology and component set have major
impact on overall costs • Cost-performance models of emulator systems are useful way to
understand the (relative) costs, test different choices• Immediate need for practical cost-performance metric(s), e.g. $/bit,
$/(Gbps/m/channel)
33
ContactsContacts
Leader: Jack Fisher (consultant)[email protected]
PCB cost model: Adam [email protected]
Optical system: Peter [email protected]
Optical materials: Bruce [email protected]
Thank YouThank YouOptoelectronics Substrates Optoelectronics Substrates
ProjectProject
34
www.inemi.orgwww.inemi.orgEmail contacts:Email contacts:
Jim McElroy Jim McElroy
[email protected]@inemi.org
Bob PfahlBob Pfahl
[email protected]@inemi.org